1 Information Structure Enric Vallduv´ıa 1.1 Introduction The term information structure is often used as a convenient cover ex- pression for a bundle of phenomena|referred to by (back)ground, com- ment, contrast, focus, given, new, rheme, theme, topic and associated terms|that exhibit pragmatic, semantic, morphosyntactic and phono- logical features. It is clear that information structure affects content, and in particular that it concerns context-sensitive aspects of content, but it is not universally agreed whether information structure forms a distinct dimension within the interpretive component of language. In this chapter, for expository purposes and as in Kruijff-Korbayov´a and Steedman (2003), the term information structure is used broadly as encompassing utterance-level features of both a semantic and a structural nature. The discussion will be centered around the following information- structural notions: the theme-rheme distinction (x1.3), givenness and topic (x1.4), and contrast (x1.5). These descriptive notions allow lin- guists to go a long way in analyzing phenomena that have generally been thought of as concerning information structure. The facts concerning fo- cus, one of the most (ab)used labels in information structure research, will be discussed in connection to the notions contrast and theme-rheme. As a first approximation to information structure consider (1) (small caps identify the lexical item associated with nuclear prominence): (1) a. We like hokey-pokey. b. Hokey-pokey we like. c. Hokey-pokey we hate. d. We like hokey-pokey. a Universitat Pompeu Fabra 2 Vallduv´ı Examples (1a) and (1b) have identical truth conditions but still differ interpretatively; there is something in the content, understood broadly, of (1a) which sets it apart from the synonymous (1b); this interpre- tive difference correlates with the difference in word order. Interestingly, (1b) and (1c) display a certain interpretative equivalence which is obvi- ously not connected to their truth-conditional meaning; this interpretive equivalence correlates with the structural `sameness' that they display. Both the interpretive difference between (1a) and (1b) and the partial in- terpretive equivalence between (1b) and (1c) are taken to be information structural in nature. There is also a difference in (non-truth-conditional) content between (1a) and (1d), which is of a nature similar to the dif- ference between (1a) and (1b); here, however, the structural contrast associated with the interpretive difference is not in the word order but rather in the intonation. In English and many other languages, if not all, intonation is an important correlate of information-structural content. The examples in (2) illustrate that the interpretive import associated with these structural differences is context dependent: (2) a. A: What are we having for dinner? b. B: We are having muttonbird for dinner. c. B0: # We are having muttonbird for dinner. d. B00: Muttonbird. Query-answer pairs are used in the literature to illustrate the connection between information structure and discourse congruence. The query in (2a) evokes a context against which uttering (2b) is felicitous whereas (2c) is not. This contrast in felicity is due to the (lack of) compati- bility between the context, as evoked by the query, and the different information-structural imports of (2b) and (2c) (the difference between them is manifested, as in the pair (1a)-(1d), via intonation). Moreover, notice that the declarative fragment in (2d), a `short-answer' counter- part of the full-fledged answer in (2b), is also felicitous in the context evoked by (2a). In fact, ceteris paribus, (2d) is a less marked answer than (2b). It is evident that context is of the essence in determining the content of (2d). That information-structural phenomena concern in one way or another the relation of utterances to the previous context appears to be the current general consensus. Thus, following Krifka and Musan (2012), information structure will be seen as referring to `those aspects of natural language that help speakers to take into consideration the addressee's Information Structure 3 current information state, and hence to facilitate the current flow of information' (2012, p. 1). This process of `facilitation' is effected with respect to a number of resources in the addressee's current information state. It is therefore necessary to adopt a particular conception of what information states are like and describe the relevant contextual resources before we actually start discussing any information-structural notions. 1.2 Contextual resources The view that context is of the essence for information structure gained wide acceptance due to the generalisation of dynamic accounts where the semantic contribution of sentences is seen as lying in their potential to change the context (see Isard, 1975). One of the main goals of the early approaches (e.g. Kamp, 1981; Heim, 1983; Seuren, 1985; Veltman, 1990; Groenendijk and Stokhof, 1991; Dekker, 1993) was to extend the analysis of a number of phenomena, like the interpretation of bound pronouns, beyond the limits of the sentence proper and into discourse. The influ- ence of earlier views on context as a set of propositions held to be true by both speaker and hearer or as a common ground of shared beliefs (e.g. Karttunen, 1974; Stalnaker, 1974) was central to these approaches, but in the drive to account for the `discourse' data, two important theoretical adjustments were made. On the one hand, it became clear that context had to be understood not as an interlocutor-neutral common ground, but rather as (the public area of) the addressee's information state at the time of utterance (tu), since there exist speaker/addressee asym- metries which can only be accounted for if each conversationalist has a distinct `dialogue gameboard' (Ginzburg, 2012). On the other hand, a subset of these early dynamic approaches felt that a view of the com- mon ground as an unstructured construct was empirically insufficient and went on to introduce a certain degree of representational richness, which structured contexts into subdomains or broadened the spectrum of contextual resources available for the computation of content. The idea that the relevant notion of context is a structured informa- tion state anchored in the addressee brought formal dynamic approaches to converge with research in other traditions, such as discourse analy- sis, (formal) pragmatics, functionalism, psycholinguistics and computa- tional linguistics, as represented, for instance, by Halliday (1967), Chafe (1976), Clark and Haviland (1977), Webber (1979), Prince (1981a), Gun- del (1985), Grosz and Sidner (1986), Sgall et al. (1986), Ariel (1988). 4 Vallduv´ı These works already elaborated in one way or another on the idea that speakers structure utterances in accordance to their assumptions about their addressee's current information states, which is the idea that in- spired representative work on information structure in the 1990s (e.g. Vallduv´ı,1992; Lambrecht, 1995; Erteschik-Shir, 1997), has survived to this day, and is reflected in the quote above by Krifka and Musan (2012).1 The erstwhile conception of context as a Stalnakerian set of com- monly agreed propositions, assumptions or facts was soon enriched with an additional resource type, namely `open propositions' (Prince, 1986), `quaestio' (Klein and von Stutterheim, 1987) or `questions under discus- sion' (quds) (Carlson, 1983; Ginzburg, 1994; Kuppevelt, 1995; Roberts, 1996; Beaver and Clark, 2008), which play a central role in the dynam- ics of dialogue and the rhetorical organisation of discourse. Quds are semantic questions (issues) which are introduced into context (at least) through uttered assertions and queries|asserting p introduces whether p (?p) and posing q introduces q|and which (potentially) constitute the subject matter of discussion and, thus, drive the progression of conversa- tion. Interlocutors must keep score of the quds that arise in conversation and also of their degree of salience: the quds in the context are partially ordered and the most salient qud is ranked highest|it is qud-maximal| at tu (Ginzburg, 2012). A qud is naturally taken to be a propositional abstract, a function from meanings of different kinds|those of the qud's potential short answers|to propositional meanings (those of the poten- tial full-fledged answers) (Tich´y,1978; von Stechow and Zimmermann, 1984; Ginzburg, 1995; Krifka, 2001).2 Quds play an essential role in ac- counting for the theme-rheme distinction in utterances. A different enrichment to contexts was the addition of a class of objects susceptible of acting as antecedents for standard pronominal anaphora 1 The origin of the terms information structure (Halliday, 1967) and information packaging (Chafe, 1976) is to be found precisely in a subset of those early works. Also essential were the ideas of the Prague School of linguistics (e.g. Sgall et al. (1986), Hajiˇcov´aet al. (1998) and references therein). Newmeyer (2001) is an informative survey of the parallelism and mutual influence between American research on information structure and the Prague School. 2 An alternative approach to quds (and to questions in general) defines the meaning of a qud as the set of (all) its possible (for some, correct) answers; see Roberts (1996) for quds and Hamblin (1973), Karttunen (1977), Groenendijk and Stokhof (1984) in general. Evidence involving information structure has been used to argue for the superiority of one or other approach (Krifka, 2001). The Presup in Jackendoff (1972) can be classed also as a qud, as it is an unsaturated propositional abstract. As for Prince's (1986) open propositions, although not formally defined as semantic questions, they are described as unsaturated abstracts and are explicitly argued to be, as a contextual resource, different from propositional assumptions or facts. Information Structure 5 and of entering into other referential links: what Karttunen (1976) called discourse referents. Discourse referents are entities of different kinds, in- cluding, at least, individual tokens, individual types, event tokens and event types.